MiniMax M3: Multimodal AI That's Surprisingly Good and Cheap

Summary

A new multimodal AI model called MiniMax M3 has been released, capable of processing text, image, and video inputs to generate text outputs. It boasts a massive one million token context window, making it suitable for complex, long-horizon tasks. While past MiniMax models haven't impressed, this new version is exceptionally fast and remarkably affordable. Currently, there's a fifty percent discount, bringing the cost down to around thirty cents for one million tokens, though the standard price is expected to triple. The model has already processed over two point two billion tokens. Early tests suggest it's incredibly efficient, with one user reporting a complex coding and design generation task costing less than sixty cents, producing output described as the best they've ever seen, even surpassing models like Opus 4.8 in design quality. The model's technical build and efficiency are highlighted as significant strengths, though some minor issues were noted. Users can access it via tools like Claude Code Router and it's also being integrated into platforms like Hermes and Pi.